While conversational agents increasingly mediate teamwork, prior work has mainly focused on when, what, or to whom an intervention is directed, with little attention to where mediation occurs. Therefore, we introduce SeeSawBot, an LLM-driven chatbot that operates across private DMs and public channels. Following a formative study, we deployed SeeSawBot in student Slack teams as a technology probe for eight weeks, collecting bi-weekly reflection surveys and post-deployment interviews. Findings show that cross-space mediation fostered sense-making across private and public spaces and redistributed emotional labor through interventions that played different relational roles over team development. We discuss cross-space mediation as both a boundary object and boundary actor, and argue that future evaluation frameworks should capture relational agency by attending to the back-and-forth negotiations through which groups construct collective understanding. We conclude with design implications that foreground where as a variable for future computational mediators, a seesaw of agency and autonomy.
ACM CHI Conference on Human Factors in Computing Systems